# pip install imbalanced-learn

Requirement already satisfied: imbalanced-learn in /Users/ulises.rey/opt/anaconda3/envs/openCV_local/lib/python3.8/site-packages (0.12.3)
Requirement already satisfied: numpy>=1.17.3 in /Users/ulises.rey/opt/anaconda3/envs/openCV_local/lib/python3.8/site-packages (from imbalanced-learn) (1.23.5)
Requirement already satisfied: scipy>=1.5.0 in /Users/ulises.rey/opt/anaconda3/envs/openCV_local/lib/python3.8/site-packages (from imbalanced-learn) (1.10.1)
Requirement already satisfied: scikit-learn>=1.0.2 in /Users/ulises.rey/opt/anaconda3/envs/openCV_local/lib/python3.8/site-packages (from imbalanced-learn) (1.2.2)
Requirement already satisfied: joblib>=1.1.1 in /Users/ulises.rey/opt/anaconda3/envs/openCV_local/lib/python3.8/site-packages (from imbalanced-learn) (1.2.0)
Requirement already satisfied: threadpoolctl>=2.0.0 in /Users/ulises.rey/opt/anaconda3/envs/openCV_local/lib/python3.8/site-packages (from imbalanced-learn) (2.2.0)
Note: you may need to restart the kernel to use updated packages.

from matplotlib import pyplot as plt
import pandas as pd
import numpy as np
import scipy.stats
from sklearn.linear_model import LinearRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import train_test_split
from sklearn.model_selection import cross_val_score
from sklearn.metrics import accuracy_score
from sklearn import ensemble
from sklearn import svm
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
import seaborn as sns
from matplotlib import pyplot as plt
from imblearn.over_sampling import SMOTE



%matplotlib inline

from google.colab import drive
drive.mount('/content/drive')

import os
os.listdir('/content/drive/My Drive/...')

import os
os.chdir('/content/drive/My Drive/...')

df = pd.read_csv('data/MI_database_final.csv')

nRow, nCol = df.shape
print(f'Hay {nRow} filas y {nCol} columnas')
df.head()

Hay 1700 filas y 104 columnas

myseed = 13

df.columns

Index(['ID', 'L_BLOOD', 'AGE', 'ALT_BLOOD', 'K_BLOOD', 'ROE', 'S_AD_ORIT',
       'AST_BLOOD', 'Na_BLOOD', 'TIME_B_S',
       ...
       'GIPER_Na', 'ritm_ecg_p_04', 'fibr_ter_03', 'GT_POST', 'fibr_ter_06',
       'np09', 'zab_leg_04', 'n_r_ecg_p_02', 'n_p_ecg_p_05', 'ZSN_A'],
      dtype='object', length=104)

X = df.drop("ZSN_A", axis=1)
y = df["ZSN_A"]

# check basic info
y.describe(), y.info()

<class 'pandas.core.series.Series'>
RangeIndex: 1700 entries, 0 to 1699
Series name: ZSN_A
Non-Null Count  Dtype  
--------------  -----  
1700 non-null   float64
dtypes: float64(1)
memory usage: 13.4 KB

(count    1700.000000
 mean        0.188235
 std         0.649065
 min         0.000000
 25%         0.000000
 50%         0.000000
 75%         0.000000
 max         4.000000
 Name: ZSN_A, dtype: float64,
 None)

# Absolute frequency
abs_freq = y.value_counts()
abs_freq

0.0    1522
1.0     103
3.0      29
2.0      27
4.0      19
Name: ZSN_A, dtype: int64

# Relative frequency
rel_freq = y.value_counts(normalize=True)
rel_freq

0.0    0.895294
1.0    0.060588
3.0    0.017059
2.0    0.015882
4.0    0.011176
Name: ZSN_A, dtype: float64

# Accumulated absolute frequency
acc_abs_freq = y.value_counts().cumsum()
acc_abs_freq

0.0    1522
1.0    1625
3.0    1654
2.0    1681
4.0    1700
Name: ZSN_A, dtype: int64

# Accumulated relative frequency
y.value_counts(normalize=True).cumsum()

0.0    0.895294
1.0    0.955882
3.0    0.972941
2.0    0.988824
4.0    1.000000
Name: ZSN_A, dtype: float64

df[["AGE", "K_BLOOD"]].mean()

AGE        61.862353
K_BLOOD     4.171471
dtype: float64

df[["AGE", "K_BLOOD"]].median()

AGE        63.0
K_BLOOD     4.1
dtype: float64

df[["AGE", "K_BLOOD"]].mode()

df["L_BLOOD"].quantile([0.15, 0.85])

0.15     5.7
0.85    11.9
Name: L_BLOOD, dtype: float64

df["L_BLOOD"].quantile([0.25, 0.5, 0.75, 1])

0.25     6.6
0.50     8.0
0.75    10.2
1.00    27.9
Name: L_BLOOD, dtype: float64

# Standard deviation
df["ROE"].std(ddof=1)

10.658735645941203

# Variance
df["ROE"].var(ddof=1)

113.60864557005763

# Measure skweness
df["AGE"].skew()

-0.22169097951636574

# Measure curtosis
df["AGE"].kurtosis()

-0.17036404643751668

df["AGE"].hist(bins=50)

<AxesSubplot:>

df["AGE"].describe()

count    1700.000000
mean       61.862353
std        11.233668
min        26.000000
25%        54.000000
50%        63.000000
75%        70.000000
max        92.000000
Name: AGE, dtype: float64

df.agg({'ALT_BLOOD': ['min', 'max', 'mean', 'std', 'var', 'skew', 'kurtosis']})

df.describe().loc[['mean']]

plt.bar(df["ZSN_A"].unique(), df["ZSN_A"].value_counts())

<BarContainer object of 5 artists>

plt.pie(df["ZSN_A"].value_counts(), labels=df["ZSN_A"].unique())

([<matplotlib.patches.Wedge at 0x7f7c18e42160>,
  <matplotlib.patches.Wedge at 0x7f7c0b32ad00>,
  <matplotlib.patches.Wedge at 0x7f7c18e428b0>,
  <matplotlib.patches.Wedge at 0x7f7c18e42d90>,
  <matplotlib.patches.Wedge at 0x7f7c18e502b0>],
 [Text(-1.0410226929209416, 0.3553473692369641, '0.0'),
  Text(0.9819462580730486, -0.4957635991642973, '1.0'),
  Text(1.0726142477705103, -0.24392350333598994, '2.0'),
  Text(1.0920737375615726, -0.13181408015950996, '3.0'),
  Text(1.099322007085675, -0.03861508432107043, '4.0')])

plt.hist(df['AGE'])

(array([  7.,  48., 104., 189., 261., 472., 314., 197.,  93.,  15.]),
 array([26. , 32.6, 39.2, 45.8, 52.4, 59. , 65.6, 72.2, 78.8, 85.4, 92. ]),
 <BarContainer object of 10 artists>)

sns.histplot(df['AGE'], kde=True)

<AxesSubplot:xlabel='AGE', ylabel='Count'>

sns.boxplot(x=df["AGE"])

<AxesSubplot:xlabel='AGE'>

columns = ["L_BLOOD", "AGE", "ALT_BLOOD","K_BLOOD","ROE","S_AD_ORIT","AST_BLOOD","Na_BLOOD","TIME_B_S"]
for column in columns:
    sns.boxplot(x=df[column])
    plt.title(f'Boxplot of {column}')  # Sets the title to the name of the current column
    plt.show()

sns.violinplot(df['AGE'])

<AxesSubplot:ylabel='AGE'>

for column in columns:
    sns.violinplot(x=df[column])
    plt.title(f'Violinplot of {column}')  # Sets the title to the name of the current column
    plt.show()

# test train split]
from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.7)

sns.histplot(data=df, x='AGE', hue='ZSN_A', stat='probability', common_norm=False, kde=True)

<AxesSubplot:xlabel='AGE', ylabel='Probability'>

# Maybe it is easier to visualize like this:
sns.histplot(data=df, x='AGE', hue='ZSN_A', stat='probability', element='step', bins=10, common_norm=False, kde=True)

<AxesSubplot:xlabel='AGE', ylabel='Probability'>

smaller_df= pd.DataFrame()

smaller_df = df[columns].copy()
smaller_df["ZSN_A"]= df['ZSN_A'].copy()

sns.pairplot(smaller_df, hue="ZSN_A")

<seaborn.axisgrid.PairGrid at 0x7f7c1b935610>

sns.scatterplot(smaller_df, x='K_BLOOD', y='L_BLOOD', hue='ZSN_A')

<AxesSubplot:xlabel='K_BLOOD', ylabel='L_BLOOD'>

# I personally would do it like this:
outcomes = smaller_df['ZSN_A'].unique()

fig, axes = plt.subplots(ncols=len(outcomes), figsize=(20, 4), sharex=True, sharey=True)
                         
for i, outcome in enumerate(outcomes):
    grouped_df = smaller_df[smaller_df['ZSN_A']==outcome]
    sns.scatterplot(grouped_df, x='K_BLOOD', y='L_BLOOD', ax=axes[i])
    axes[i].set_title(f"Outcome of ZSN_A is: {outcome}")

# Or same but with a reglplot
outcomes = smaller_df['ZSN_A'].unique()

fig, axes = plt.subplots(ncols=len(outcomes), figsize=(25, 4), sharex=True, sharey=True)
                         
for i, outcome in enumerate(outcomes):
    grouped_df = smaller_df[smaller_df['ZSN_A']==outcome]
    sns.regplot(grouped_df, x='K_BLOOD', y='L_BLOOD', ax=axes[i])
    axes[i].set_title(f"Outcome of ZSN_A is: {outcome}")

sns.scatterplot(smaller_df, x='ROE', y='AST_BLOOD', hue='ZSN_A')

<AxesSubplot:xlabel='ROE', ylabel='AST_BLOOD'>

# Or same but with a reglplot
outcomes = smaller_df['ZSN_A'].unique()

fig, axes = plt.subplots(ncols=len(outcomes), figsize=(20, 4), sharex=True, sharey=True)
                         
for i, outcome in enumerate(outcomes):
    grouped_df = smaller_df[smaller_df['ZSN_A']==outcome]
    sns.regplot(grouped_df, x='ROE', y='AST_BLOOD', ax=axes[i])
    axes[i].set_title(f"Outcome of ZSN_A is: {outcome}")
    axes[i].set_ylim([0, 2.2])

df.corr(method='pearson')

df.corr(method='spearman')

df.corr(method='kendall')

def histogram_intersection(var1, var2):
    """
    Return the histogram intersection of two variables. Does not handle Nan.
    """
    hist_1, _ = np.histogram(var1, bins=100, range=[-15, 15])
    hist_2, _ = np.histogram(var2, bins=100, range=[-15, 15])
    minima = np.minimum(hist_1, hist_2)
    intersection = np.true_divide(np.sum(minima), np.sum(hist_2))
    return intersection

histogram_intersection(df['L_BLOOD'], df['ALT_BLOOD'])

0.002352941176470588

sns.heatmap(df.corr(method='pearson'))

<AxesSubplot:>

sns.heatmap(smaller_df.corr(method='pearson'))

<AxesSubplot:>

# test train split]
from sklearn.model_selection import train_test_split

# It says test should be 0.7, but the teacher wrote on the forum that is an error, it should be train 0.7

X_train, X_test, y_train, y_test = train_test_split(X, y, train_size=0.7)

X.columns

Index(['ID', 'L_BLOOD', 'AGE', 'ALT_BLOOD', 'K_BLOOD', 'ROE', 'S_AD_ORIT',
       'AST_BLOOD', 'Na_BLOOD', 'TIME_B_S',
       ...
       'fibr_ter_01', 'GIPER_Na', 'ritm_ecg_p_04', 'fibr_ter_03', 'GT_POST',
       'fibr_ter_06', 'np09', 'zab_leg_04', 'n_r_ecg_p_02', 'n_p_ecg_p_05'],
      dtype='object', length=103)

y_train.unique()

array([0., 2., 1., 3., 4.])

len(y_train), y_train.value_counts(normalize=True)

(1190,
 0.0    0.903361
 1.0    0.056303
 3.0    0.016807
 2.0    0.012605
 4.0    0.010924
 Name: ZSN_A, dtype: float64)

# Smote
smote = SMOTE(random_state=myseed, k_neighbors=2)
X_train_smote, y_train_smote = smote.fit_resample(X_train, y_train)

len(y_train_smote), y_train_smote.value_counts(normalize=True)

(5375,
 0.0    0.2
 2.0    0.2
 1.0    0.2
 3.0    0.2
 4.0    0.2
 Name: ZSN_A, dtype: float64)

scaler = StandardScaler()
X_train_smote_scaled = scaler.fit_transform(X_train_smote)
X_test_scaled = scaler.transform(X_test)

dtree = DecisionTreeClassifier(max_depth=3, random_state=myseed)

scores = cross_val_score(dtree, X_train_smote_scaled, y_train_smote, cv=5)
scores

array([0.59906977, 0.58976744, 0.61488372, 0.62046512, 0.60651163])

#If we want to test on the test split
dtree.fit(X_train_smote_scaled, y_train_smote)

DecisionTreeClassifier(max_depth=3, random_state=13)

DecisionTreeClassifier(max_depth=3, random_state=13)

y_pred = dtree.predict(X_test_scaled)

accuracy_score(y_test, y_pred)

0.5666666666666667

def train_and_predict(model):
    """
    Train and predict the current dataset based on a given model
    """
    model.fit(X_train_smote_scaled, y_train_smote)
    y_pred = model.predict(X_test_scaled)
    acc_score = accuracy_score(y_pred, y_test)
    
    #regex to get the name of the model from the class
    model_name = str(type(model)).split(".")[-1][:-2]
    
    print(f"With model {model_name}, we have achieved an accuracy score of {acc_score:.3f} on the test data.")
    return {model_name:acc_score}

# We verify it works
dtree_acc = train_and_predict(dtree)

With model DecisionTreeClassifier, we have achieved an accuracy score of 0.567 on the test data.

from sklearn.ensemble import RandomForestClassifier

rand_forest = RandomForestClassifier()

cross_val_score(rand_forest, X_train_smote_scaled, y_train_smote, cv=5)

array([0.9172093 , 1.        , 0.99906977, 1.        , 1.        ])

rand_forest_acc = train_and_predict(rand_forest)

With model RandomForestClassifier, we have achieved an accuracy score of 0.876 on the test data.

from sklearn.ensemble import GradientBoostingClassifier
grad_boost = GradientBoostingClassifier(n_estimators=20, random_state=myseed, max_depth=3)

cross_val_score(grad_boost, X_train_smote_scaled, y_train_smote, cv=5)

array([0.92465116, 0.9655814 , 0.96465116, 0.97860465, 0.97488372])

grad_boost_acc = train_and_predict(grad_boost)

With model GradientBoostingClassifier, we have achieved an accuracy score of 0.824 on the test data.

comb_dict = {**dtree_acc, **rand_forest_acc, **grad_boost_acc}

acc_df = pd.DataFrame(list(comb_dict.items()), columns=['Model', 'Accuracy'])

acc_df

sns.barplot(x='Model', y='Accuracy', data=acc_df)
plt.xticks(rotation=30)

([0, 1, 2], [Text(0, 0, ''), Text(0, 0, ''), Text(0, 0, '')])

knn_classifier = KNeighborsClassifier(n_neighbors=2)

cross_val_score(knn_classifier, X_train_smote_scaled, y_train_smote, cv=5)

array([0.96372093, 0.97116279, 0.97116279, 0.97209302, 0.96744186])

knn_acc = train_and_predict(knn_classifier)

With model KNeighborsClassifier, we have achieved an accuracy score of 0.753 on the test data.

from sklearn.svm import SVC
gamma=0.07
svm = SVC(gamma=gamma, random_state=myseed)

cross_val_score(svm, X_train_smote_scaled, y_train_smote, cv=5)

array([0.99627907, 0.99813953, 0.99627907, 0.9972093 , 0.9972093 ])

svm_acc = train_and_predict(svm)

With model SVC, we have achieved an accuracy score of 0.876 on the test data.

from sklearn.ensemble import StackingClassifier

#final model
grad_boost = GradientBoostingClassifier(n_estimators=20, max_depth=3, random_state=myseed)

# models to stack
dtree = DecisionTreeClassifier(max_depth=3, random_state=myseed)
knn_classifier = KNeighborsClassifier(n_neighbors=2)
gamma=0.07
svm = SVC(gamma=gamma, random_state=myseed)

estimators = [
    ("decision_tree", dtree),
     ("knn_classifier", knn_classifier),
     ("svm", svm)]

# Stacked classifier
stacked_classifier = StackingClassifier(estimators=estimators, final_estimator=grad_boost, passthrough=False)

cross_val_score(stacked_classifier, X_train_smote_scaled, y_train_smote)

array([0.9972093 , 0.99813953, 1.        , 0.99906977, 0.99906977])

stack_acc = train_and_predict(stacked_classifier)

With model StackingClassifier, we have achieved an accuracy score of 0.876 on the test data.

#final model
grad_boost = GradientBoostingClassifier(n_estimators=20, max_depth=3, random_state=myseed)

# models to stack
dtree = DecisionTreeClassifier(max_depth=3, random_state=myseed)
knn_classifier = KNeighborsClassifier(n_neighbors=2)
gamma=0.07
svm = SVC(gamma=gamma, random_state=myseed)

estimators = [
    ("decision_tree", dtree),
     ("knn_classifier", knn_classifier),
     ("svm", svm)]

# stacked classifier WITH pasthrough of orignal data set to TRUE
stacked_classifier_passthrough = StackingClassifier(estimators=estimators, final_estimator=grad_boost, passthrough=True)

cross_val_score(stacked_classifier_passthrough, X_train_smote_scaled, y_train_smote)

array([0.99906977, 0.99813953, 0.99906977, 0.99906977, 1.        ])

stack_acc_pt = train_and_predict(stacked_classifier_passthrough)

With model StackingClassifier, we have achieved an accuracy score of 0.876 on the test data.

stack_acc_pt

{'StackingClassifier': 0.8764705882352941}

# Change its name (key) on the dict
stack_acc_pt['StackingClassifierCascade'] = stack_acc_pt.pop('StackingClassifier')

comb_dict = {**dtree_acc, **rand_forest_acc, **grad_boost_acc, **stack_acc, **stack_acc_pt}
acc_df = pd.DataFrame(list(comb_dict.items()), columns=['Model', 'Accuracy'])
acc_df

sns.barplot(x='Model', y='Accuracy', data=acc_df)
plt.xticks(rotation=30)

([0, 1, 2, 3, 4],
 [Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, ''),
  Text(0, 0, '')])

# !pip install numpy pandas scikit-learn tensorflow keras

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler, OneHotEncoder
from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay
from keras.models import Model
from keras.layers import Input, Dense, concatenate
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense
from tensorflow.keras.layers import Input, concatenate
from tensorflow.keras.models import Model

# y_train_balanced and y_test are integer class labels
y_train_balanced_encoded = to_categorical(y_train_smote)
y_test_encoded = to_categorical(y_test)

from tensorflow.keras.optimizers import Adam

from tensorflow.keras.utils import plot_model

# Definir el model seqüencial
model_seq = Sequential()

# Capa d'entrada
input_dim = X_train_smote_scaled.shape[1]

# Primera capa oculta
model_seq.add(Dense(64, activation='relu', input_dim=input_dim))

# Segona capa oculta
model_seq.add(Dense(64, activation='relu'))

# Capa de sortida
output_dim = y_train_balanced_encoded.shape[1]
model_seq.add(Dense(output_dim, activation='softmax'))

# Compilar el model
model_seq.compile(optimizer=Adam(), loss='categorical_crossentropy', metrics=['accuracy'])

model_seq.summary()

Model: "sequential_4"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 dense_91 (Dense)            (None, 64)                6656      
                                                                 
 dense_92 (Dense)            (None, 64)                4160      
                                                                 
 dense_93 (Dense)            (None, 5)                 325       
                                                                 
=================================================================
Total params: 11141 (43.52 KB)
Trainable params: 11141 (43.52 KB)
Non-trainable params: 0 (0.00 Byte)
_________________________________________________________________

plot_model(model_seq)

# Train
history = model_seq.fit(X_train_smote_scaled, y_train_balanced_encoded, epochs=50, batch_size=10, verbose=1)

Epoch 1/50
538/538 [==============================] - 2s 2ms/step - loss: 0.3730 - accuracy: 0.8856
Epoch 2/50
538/538 [==============================] - 1s 2ms/step - loss: 0.0596 - accuracy: 0.9855
Epoch 3/50
538/538 [==============================] - 1s 2ms/step - loss: 0.0218 - accuracy: 0.9953
Epoch 4/50
538/538 [==============================] - 1s 2ms/step - loss: 0.0091 - accuracy: 0.9989
Epoch 5/50
538/538 [==============================] - 1s 2ms/step - loss: 0.0039 - accuracy: 0.9996
Epoch 6/50
538/538 [==============================] - 1s 2ms/step - loss: 0.0016 - accuracy: 1.0000
Epoch 7/50
538/538 [==============================] - 1s 2ms/step - loss: 8.3265e-04 - accuracy: 1.0000
Epoch 8/50
538/538 [==============================] - 1s 2ms/step - loss: 5.0491e-04 - accuracy: 1.0000
Epoch 9/50
538/538 [==============================] - 1s 2ms/step - loss: 3.2983e-04 - accuracy: 1.0000
Epoch 10/50
538/538 [==============================] - 1s 2ms/step - loss: 2.2278e-04 - accuracy: 1.0000
Epoch 11/50
538/538 [==============================] - 1s 2ms/step - loss: 1.5219e-04 - accuracy: 1.0000
Epoch 12/50
538/538 [==============================] - 1s 2ms/step - loss: 1.0734e-04 - accuracy: 1.0000
Epoch 13/50
538/538 [==============================] - 1s 2ms/step - loss: 7.5852e-05 - accuracy: 1.0000
Epoch 14/50
538/538 [==============================] - 1s 2ms/step - loss: 5.4631e-05 - accuracy: 1.0000
Epoch 15/50
538/538 [==============================] - 1s 1ms/step - loss: 3.8788e-05 - accuracy: 1.0000
Epoch 16/50
538/538 [==============================] - 1s 1ms/step - loss: 2.8099e-05 - accuracy: 1.0000
Epoch 17/50
538/538 [==============================] - 1s 2ms/step - loss: 1.9902e-05 - accuracy: 1.0000
Epoch 18/50
538/538 [==============================] - 1s 2ms/step - loss: 1.4556e-05 - accuracy: 1.0000
Epoch 19/50
538/538 [==============================] - 1s 2ms/step - loss: 1.0615e-05 - accuracy: 1.0000
Epoch 20/50
538/538 [==============================] - 1s 2ms/step - loss: 7.6223e-06 - accuracy: 1.0000
Epoch 21/50
538/538 [==============================] - 1s 2ms/step - loss: 5.5650e-06 - accuracy: 1.0000
Epoch 22/50
538/538 [==============================] - 1s 2ms/step - loss: 4.0371e-06 - accuracy: 1.0000
Epoch 23/50
538/538 [==============================] - 1s 2ms/step - loss: 2.9962e-06 - accuracy: 1.0000
Epoch 24/50
538/538 [==============================] - 1s 2ms/step - loss: 2.1669e-06 - accuracy: 1.0000
Epoch 25/50
538/538 [==============================] - 1s 2ms/step - loss: 1.6033e-06 - accuracy: 1.0000
Epoch 26/50
538/538 [==============================] - 1s 2ms/step - loss: 1.1786e-06 - accuracy: 1.0000
Epoch 27/50
538/538 [==============================] - 1s 2ms/step - loss: 8.7157e-07 - accuracy: 1.0000
Epoch 28/50
538/538 [==============================] - 1s 2ms/step - loss: 6.3113e-07 - accuracy: 1.0000
Epoch 29/50
538/538 [==============================] - 1s 2ms/step - loss: 4.6530e-07 - accuracy: 1.0000
Epoch 30/50
538/538 [==============================] - 1s 2ms/step - loss: 3.4346e-07 - accuracy: 1.0000
Epoch 31/50
538/538 [==============================] - 1s 2ms/step - loss: 2.5193e-07 - accuracy: 1.0000
Epoch 32/50
538/538 [==============================] - 1s 2ms/step - loss: 1.8712e-07 - accuracy: 1.0000
Epoch 33/50
538/538 [==============================] - 1s 2ms/step - loss: 1.3666e-07 - accuracy: 1.0000
Epoch 34/50
538/538 [==============================] - 1s 1ms/step - loss: 1.0089e-07 - accuracy: 1.0000
Epoch 35/50
538/538 [==============================] - 1s 2ms/step - loss: 7.4054e-08 - accuracy: 1.0000
Epoch 36/50
538/538 [==============================] - 1s 2ms/step - loss: 5.4115e-08 - accuracy: 1.0000
Epoch 37/50
538/538 [==============================] - 1s 2ms/step - loss: 4.0365e-08 - accuracy: 1.0000
Epoch 38/50
538/538 [==============================] - 1s 1ms/step - loss: 2.9586e-08 - accuracy: 1.0000
Epoch 39/50
538/538 [==============================] - 1s 1ms/step - loss: 2.2378e-08 - accuracy: 1.0000
Epoch 40/50
538/538 [==============================] - 1s 1ms/step - loss: 1.6523e-08 - accuracy: 1.0000
Epoch 41/50
538/538 [==============================] - 1s 1ms/step - loss: 1.2220e-08 - accuracy: 1.0000
Epoch 42/50
538/538 [==============================] - 1s 1ms/step - loss: 9.1154e-09 - accuracy: 1.0000
Epoch 43/50
538/538 [==============================] - 1s 2ms/step - loss: 6.8310e-09 - accuracy: 1.0000
Epoch 44/50
538/538 [==============================] - 1s 2ms/step - loss: 4.9458e-09 - accuracy: 1.0000
Epoch 45/50
538/538 [==============================] - 2s 4ms/step - loss: 4.1252e-09 - accuracy: 1.0000
Epoch 46/50
538/538 [==============================] - 1s 2ms/step - loss: 3.1050e-09 - accuracy: 1.0000
Epoch 47/50
538/538 [==============================] - 1s 2ms/step - loss: 2.1513e-09 - accuracy: 1.0000
Epoch 48/50
538/538 [==============================] - 1s 1ms/step - loss: 1.7299e-09 - accuracy: 1.0000
Epoch 49/50
538/538 [==============================] - 1s 2ms/step - loss: 1.2642e-09 - accuracy: 1.0000
Epoch 50/50
538/538 [==============================] - 1s 2ms/step - loss: 1.0867e-09 - accuracy: 1.0000

def training_plot(history):
    """
    Plot the training loss and accuracy, next to the validation loss and accuracy
    Params: history object from training a keras model
    Return: Fig, axes
    """
    fig, axes = plt.subplots(ncols=2, figsize=(12, 4))
    axes[0].plot(history.history['accuracy'], label='Training')
    #axes[0].plot(history.history['val_accuracy'], label='Validation')
    axes[0].set_title('Accuracy')
    axes[0].legend()
    axes[0].set_ylabel("Accuracy")
    axes[1].plot(history.history['loss'], label='Training')
    #axes[1].plot(history.history['val_loss'], label='Validation')
    axes[1].set_title('Loss')
    axes[1].legend()
    axes[1].set_ylabel("Loss")
    for ax in axes:
        ax.set_xlabel("Epoch")
    plt.show()

    return fig, axes

training_plot(history)

(<Figure size 1200x400 with 2 Axes>,
 array([<AxesSubplot:title={'center':'Accuracy'}, xlabel='Epoch', ylabel='Accuracy'>,
        <AxesSubplot:title={'center':'Loss'}, xlabel='Epoch', ylabel='Loss'>],
       dtype=object))

y_pred_prob_seq = model_seq.predict(X_test_scaled)
y_pred_prob_seq

16/16 [==============================] - 0s 1ms/step

array([[9.9999982e-01, 4.7393687e-08, 1.7201629e-08, 9.6744913e-08,
        1.6019825e-09],
       [9.9999994e-01, 2.6907721e-11, 7.5566484e-20, 3.9805545e-13,
        3.7622987e-17],
       [9.9999994e-01, 2.2662623e-12, 1.2833130e-11, 3.4072207e-11,
        7.1539598e-16],
       ...,
       [1.0000000e+00, 2.8214384e-11, 4.7209046e-18, 2.7665483e-16,
        1.9700219e-20],
       [1.0000000e+00, 6.4241577e-11, 4.3006118e-19, 1.2123955e-19,
        2.6655768e-20],
       [9.9840063e-01, 5.0945108e-08, 3.0140429e-06, 1.1070739e-15,
        1.5963641e-03]], dtype=float32)

y_pred_seq = np.argmax(y_pred_prob_seq, axis=1)
y_pred_seq

array([0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       1, 0, 0, 0, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 3, 0, 0, 3,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 3, 1, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 1, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 3, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 4, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
       0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
       0, 0, 0, 0])

# ground truth vs pred 
plt.scatter(y_test, y_pred_seq)
plt.xlabel("Ground Truth Labels")
plt.ylabel("Predicted Labels")

Text(0, 0.5, 'Predicted Labels')

acc_score = accuracy_score(y_pred_seq, y_test)
    
# Extract the name of the model from the class
model_name = str(type(model_seq)).split(".")[-1][:-2]
    
print(f"With model {model_name}, we have achieved an accuracy score of {acc_score:.3f} on the test data.")

With model Sequential, we have achieved an accuracy score of 0.841 on the test data.

# Calculate the correct split size to ensure even feature distribution for the parallel inputs
split_size = X_test_scaled.shape[1] // 2
if X_test_scaled.shape[1] % 2 != 0:  # Adjust if the total number of features is odd
   split_size += 1

from tensorflow.keras.layers import Input, Dense, Dropout, concatenate

def train_parallel_model(n_neurons):
    split_size = X_train_smote_scaled.shape[1] // 2

    if X_train_smote_scaled.shape[1] % 2 != 0:
        split_size += 1

    # Definir les capes d'entrada
    input_a = Input(shape=(split_size,))
    input_b = Input(shape=(X_train_smote_scaled.shape[1] - split_size,))

    # Ruta A
    x = Dense(n_neurons, activation='relu')(input_a)
    x = Dropout(0.2)(x)

    # Ruta B
    y = Dense(n_neurons, activation='relu')(input_b)
    y = Dropout(0.2)(y)

    # Concatenació i capes finals
    merged = concatenate([x, y])
    z = Dense(64, activation='relu')(merged)
    output = Dense(y_train_balanced_encoded.shape[1], activation='softmax')(z)

    # Crear el model
    model = Model(inputs=[input_a, input_b], outputs=output)

    # Compilar el model
    model.compile(optimizer=Adam(), loss='categorical_crossentropy', metrics=['accuracy'])
    
    # Entrenar el model
    history = model.fit([X_train_smote_scaled[:, :split_size], X_train_smote_scaled[:, split_size:]],
                        y_train_balanced_encoded, epochs=10, batch_size=64, validation_split=0.1, verbose=0)

    # Avaluar el model
    y_pred_prob = model.predict([X_test_scaled[:, :split_size], X_test_scaled[:, split_size:]])
    y_pred = np.argmax(y_pred_prob, axis=1)

    acc_score = accuracy_score(y_test, y_pred)
    
    return model, y_pred, acc_score

results = {}
for n_neurons in range(30, 71, 10):
    model, y_pred, acc_score = train_parallel_model(n_neurons)
    results[n_neurons] = acc_score
    print(f"With {n_neurons} neurons, achieved accuracy: {acc_score:.3f}")

16/16 [==============================] - 0s 2ms/step
With 30 neurons, achieved accuracy: 0.747
16/16 [==============================] - 0s 2ms/step
With 40 neurons, achieved accuracy: 0.818
16/16 [==============================] - 0s 2ms/step
With 50 neurons, achieved accuracy: 0.822
16/16 [==============================] - 0s 1ms/step
With 60 neurons, achieved accuracy: 0.829
16/16 [==============================] - 0s 1ms/step
With 70 neurons, achieved accuracy: 0.829

model, y_pred_parallel, acc_score = train_parallel_model(70)

16/16 [==============================] - 0s 2ms/step

acc_score = accuracy_score(y_test,y_pred_parallel)
print(f"With model parallel, we have achieved an accuracy score of {acc_score:.3f} on the test data.")

With model parallel, we have achieved an accuracy score of 0.831 on the test data.

# Function to visualize the confusion matrix
def plot_confusion_matrix(cm, class_names, fmt=".2f"):
    """
    Plot the confusion matrix calculated from sklearn.confusion_matrix()

    """
    plt.figure(figsize=(10, 10))

    sns.heatmap(cm, annot=True, fmt=fmt, cmap='Blues', xticklabels=class_names, yticklabels=class_names)
    plt.xlabel('Predicted label')
    plt.ylabel('True label')
    plt.show()

# Compute confusion matrix

cm = confusion_matrix(y_test,y_pred_seq)

# Compute the normalized confusion matrix

cm_norm = confusion_matrix(y_test,y_pred_seq, normalize="true")

class_names = ['Class0', 'Class1', 'Class2', 'Class3', 'Class4']

plot_confusion_matrix(cm_norm, class_names, fmt=".2f")

cm = confusion_matrix(y_test,y_pred_parallel)

# Compute the normalized confusion matrix

cm_norm = confusion_matrix(y_test,y_pred_parallel, normalize="true")

plot_confusion_matrix(cm_norm, class_names, fmt=".2f")

	ID	L_BLOOD	AGE	ALT_BLOOD	K_BLOOD	ROE	S_AD_ORIT	AST_BLOOD	Na_BLOOD	TIME_B_S	...	GIPER_Na	ritm_ecg_p_04	fibr_ter_03	GT_POST	fibr_ter_06	np09	zab_leg_04	n_r_ecg_p_02	n_p_ecg_p_05	ZSN_A
ID	1.000000	0.165842	0.213113	-0.063048	-0.054767	0.066894	-0.155869	0.002653	-0.027627	-0.116248	...	-0.014430	0.096467	0.086920	0.036185	-0.025693	-0.053010	0.036512	0.029687	0.034128	0.047305
L_BLOOD	0.165842	1.000000	0.001554	0.044013	0.014832	0.008114	-0.114370	0.075467	0.014114	-0.102946	...	0.017224	0.002667	0.046766	0.015199	0.031578	-0.042138	0.018969	0.036172	0.017531	0.048912
AGE	0.213113	0.001554	1.000000	-0.100174	-0.006518	0.197549	0.045591	-0.052900	0.025094	-0.047046	...	-0.028592	0.030906	-0.169636	0.048295	-0.069120	-0.019443	0.016052	0.029926	0.040149	0.120281
ALT_BLOOD	-0.063048	0.044013	-0.100174	1.000000	0.019434	-0.009963	-0.079810	0.522017	-0.009438	0.019828	...	0.055006	-0.017821	-0.022610	0.006627	0.012534	0.020331	-0.005944	-0.011515	-0.008162	0.009486
K_BLOOD	-0.054767	0.014832	-0.006518	0.019434	1.000000	0.012192	0.018944	0.046188	0.300705	0.080468	...	0.158989	0.026362	-0.032646	0.023540	-0.021168	0.029743	0.063832	-0.004787	-0.003674	0.020274
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
np09	-0.053010	-0.042138	-0.019443	0.020331	0.029743	-0.016212	-0.004606	0.008101	-0.002562	-0.014057	...	-0.004600	-0.004019	-0.007006	-0.002360	-0.002504	1.000000	-0.002504	-0.002360	-0.001178	-0.009956
zab_leg_04	0.036512	0.018969	0.016052	-0.005944	0.063832	0.030960	-0.012603	-0.010419	-0.016722	-0.007913	...	-0.009778	-0.008544	-0.014892	-0.005016	-0.005322	-0.002504	1.000000	-0.005016	-0.002504	-0.021164
n_r_ecg_p_02	0.029687	0.036172	0.029926	-0.011515	-0.004787	0.019949	-0.016684	-0.009355	-0.026055	0.024634	...	-0.009216	-0.008053	-0.014036	-0.004728	-0.005016	-0.002360	-0.005016	1.000000	-0.002360	-0.019947
n_p_ecg_p_05	0.034128	0.017531	0.040149	-0.008162	-0.003674	0.007944	-0.046285	-0.006759	-0.002562	0.004545	...	-0.004600	-0.004019	-0.007006	-0.002360	-0.002504	-0.001178	-0.002504	-0.002360	1.000000	-0.009956
ZSN_A	0.047305	0.048912	0.120281	0.009486	0.020274	0.021462	-0.042863	-0.001787	-0.018979	-0.007118	...	-0.025111	-0.033973	-0.059215	-0.019947	-0.021164	-0.009956	-0.021164	-0.019947	-0.009956	1.000000

	ID	L_BLOOD	AGE	ALT_BLOOD	K_BLOOD	ROE	S_AD_ORIT	AST_BLOOD	Na_BLOOD	TIME_B_S	...	GIPER_Na	ritm_ecg_p_04	fibr_ter_03	GT_POST	fibr_ter_06	np09	zab_leg_04	n_r_ecg_p_02	n_p_ecg_p_05	ZSN_A
ID	1.000000	0.174530	0.213500	-0.070312	-0.055582	0.063879	-0.115106	0.001847	-0.046077	-0.112542	...	-0.014430	0.096467	0.086920	0.036185	-0.025693	-0.053010	0.036512	0.029687	0.034128	-0.079351
L_BLOOD	0.174530	1.000000	0.012622	0.010579	0.008495	-0.008840	-0.097681	0.040632	-0.007923	-0.103573	...	0.007732	0.019099	0.056602	0.016628	0.031298	-0.052158	-0.003363	0.036244	0.023351	0.038419
AGE	0.213500	0.012622	1.000000	-0.115943	-0.018558	0.212178	0.077114	-0.074974	0.007550	-0.034547	...	-0.030767	0.035252	-0.165470	0.052734	-0.068530	-0.007906	0.015881	0.029692	0.044149	0.121550
ALT_BLOOD	-0.070312	0.010579	-0.115943	1.000000	0.021411	-0.009946	-0.067855	0.509998	0.012675	0.025233	...	0.032465	-0.003894	-0.021003	0.014086	0.020561	0.035100	0.001919	-0.000301	0.000141	-0.004886
K_BLOOD	-0.055582	0.008495	-0.018558	0.021411	1.000000	-0.002136	0.010957	0.038051	0.287098	0.097939	...	0.123672	0.001345	-0.045503	0.016990	-0.018764	0.027076	0.043353	0.006273	0.000106	0.011426
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
np09	-0.053010	-0.052158	-0.007906	0.035100	0.027076	-0.014040	-0.005297	0.008694	-0.004827	-0.007593	...	-0.004600	-0.004019	-0.007006	-0.002360	-0.002504	1.000000	-0.002504	-0.002360	-0.001178	-0.011718
zab_leg_04	0.036512	-0.003363	0.015881	0.001919	0.043353	0.023735	-0.009362	0.006959	-0.003150	-0.002844	...	-0.009778	-0.008544	-0.014892	-0.005016	-0.005322	-0.002504	1.000000	-0.005016	-0.002504	-0.024908
n_r_ecg_p_02	0.029687	0.036244	0.029692	-0.000301	0.006273	0.022759	-0.017905	-0.004896	-0.035830	0.023324	...	-0.009216	-0.008053	-0.014036	-0.004728	-0.005016	-0.002360	-0.005016	1.000000	-0.002360	-0.023477
n_p_ecg_p_05	0.034128	0.023351	0.044149	0.000141	0.000106	0.019919	-0.031925	0.003612	-0.004827	-0.001254	...	-0.004600	-0.004019	-0.007006	-0.002360	-0.002504	-0.001178	-0.002504	-0.002360	1.000000	-0.011718
ZSN_A	-0.079351	0.038419	0.121550	-0.004886	0.011426	0.053387	-0.015079	-0.030352	-0.017657	-0.002250	...	-0.017910	-0.039985	-0.069693	-0.023477	-0.024908	-0.011718	-0.024908	-0.023477	-0.011718	1.000000

	ID	L_BLOOD	AGE	ALT_BLOOD	K_BLOOD	ROE	S_AD_ORIT	AST_BLOOD	Na_BLOOD	TIME_B_S	...	GIPER_Na	ritm_ecg_p_04	fibr_ter_03	GT_POST	fibr_ter_06	np09	zab_leg_04	n_r_ecg_p_02	n_p_ecg_p_05	ZSN_A
ID	1.000000	0.116703	0.144797	-0.048414	-0.038322	0.043893	-0.080464	0.001192	-0.030603	-0.080973	...	-0.011785	0.078788	0.070991	0.029554	-0.020985	-0.043295	0.029821	0.024246	0.027873	-0.065649
L_BLOOD	0.116703	1.000000	0.008352	0.007379	0.005844	-0.006119	-0.069245	0.028862	-0.005574	-0.073273	...	0.006366	0.015724	0.046598	0.013690	0.025766	-0.042940	-0.002768	0.029839	0.019224	0.031109
AGE	0.144797	0.008352	1.000000	-0.082974	-0.012769	0.148083	0.054910	-0.053246	0.005201	-0.025058	...	-0.025457	0.029168	-0.136913	0.043633	-0.056704	-0.006542	0.013140	0.024568	0.036530	0.098885
ALT_BLOOD	-0.048414	0.007379	-0.082974	1.000000	0.015101	-0.007227	-0.050956	0.409324	0.010793	0.018627	...	0.027920	-0.003349	-0.018063	0.012114	0.017683	0.030186	0.001651	-0.000259	0.000122	-0.004121
K_BLOOD	-0.038322	0.005844	-0.012769	0.015101	1.000000	-0.000513	0.007976	0.028866	0.218048	0.071913	...	0.104905	0.001141	-0.038598	0.014412	-0.015917	0.022967	0.036774	0.005321	0.000090	0.009588
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
np09	-0.043295	-0.042940	-0.006542	0.030186	0.022967	-0.011806	-0.004599	0.007499	-0.004181	-0.006585	...	-0.004600	-0.004019	-0.007006	-0.002360	-0.002504	1.000000	-0.002504	-0.002360	-0.001178	-0.011535
zab_leg_04	0.029821	-0.002768	0.013140	0.001651	0.036774	0.019958	-0.008129	0.006002	-0.002728	-0.002467	...	-0.009778	-0.008544	-0.014892	-0.005016	-0.005322	-0.002504	1.000000	-0.005016	-0.002504	-0.024520
n_r_ecg_p_02	0.024246	0.029839	0.024568	-0.000259	0.005321	0.019137	-0.015546	-0.004223	-0.031033	0.020226	...	-0.009216	-0.008053	-0.014036	-0.004728	-0.005016	-0.002360	-0.005016	1.000000	-0.002360	-0.023110
n_p_ecg_p_05	0.027873	0.019224	0.036530	0.000122	0.000090	0.016749	-0.027719	0.003116	-0.004181	-0.001087	...	-0.004600	-0.004019	-0.007006	-0.002360	-0.002504	-0.001178	-0.002504	-0.002360	1.000000	-0.011535
ZSN_A	-0.065649	0.031109	0.098885	-0.004121	0.009588	0.044056	-0.012937	-0.025724	-0.015136	-0.001917	...	-0.017631	-0.039360	-0.068605	-0.023110	-0.024520	-0.011535	-0.024520	-0.023110	-0.011535	1.000000

	ID	L_BLOOD	AGE	ALT_BLOOD	K_BLOOD	ROE	S_AD_ORIT	AST_BLOOD	Na_BLOOD	TIME_B_S	...	ZSN_A
0	1	8.0	77.0	0.38	4.7	16.0	180.0	0.22	138.0	4.0	...	0.0
1	2	7.8	55.0	0.38	3.5	3.0	120.0	0.18	132.0	2.0	...	0.0
2	3	10.8	52.0	0.30	4.0	10.0	180.0	0.11	132.0	3.0	...	0.0
3	4	8.0	68.0	0.75	3.9	10.0	120.0	0.37	146.0	2.0	...	1.0
4	5	8.3	60.0	0.45	3.5	10.0	160.0	0.22	132.0	9.0	...	0.0

	ALT_BLOOD
min	0.030000
max	3.000000
mean	0.464506
std	0.355437
var	0.126336
skew	2.583580
kurtosis	9.172009

	Model	Accuracy
0	DecisionTreeClassifier	0.566667
1	RandomForestClassifier	0.876471
2	GradientBoostingClassifier	0.823529

PAC 4: Combinació de models de classificació¶

Objectiu:¶

Requisits:¶

Anàlisi estadística per identificar el desequilibri:¶

Divisió i estandardització de dades:¶

Construcció de models i ajustament d'hiperparàmetres:¶

Avaluació del model:¶

Consideracions generals:¶

Format del lliurament:¶

Infart de Miocardi (IM):¶

Ruta de treball a Colab¶

1. Càrrega i ajustament de la base de dades MI (1.5 punts)¶

1.1 Dades¶

1.2 Anàlisi descriptiva univariable¶

Taula de freqüència¶

1.3 Mesures de tendència central: mitjana, mediana i moda¶

1.4 Mesures de posició: quartils, decils i percentils¶

1.5 Mesures de dispersió: desviació típica, rang, IQR, variància¶

1.6 Mesures d'asimetria (skewness) i apuntalament (curtosi)¶

1.7 Resum de totes les estadístiques¶

1.8 Anàlisi gràfica univariable¶

1.8.1 Gràfic de barres¶

1.8.2 Diagrama de sectors¶

1.8.3 Histograma¶

1.8.4 Estimació de densitat de Kernel¶

1.8.5 Diagrama de caixes¶

1.8.6 Diagrama de violí¶

1.9 Anàlisi bivariable¶

Coeficients de correlació¶

2. Combinació de classificadors (5 punts)¶

2.1. Combinació paral·lela de classificadors (2.5 punts)¶

2.1.1. Arbres de decisió¶

2.1.2. Bagging¶

2.1.3. Boosting¶

2.1.4. Anàlisi¶

2.2. Combinació seqüencial de classificadors base diferents (2.5 punts)¶

2.2.1. KNN (k veïns més propers)¶

2.2.2. SVM (Support Vector Machines)¶

2.2.3. Stacking¶

2.2.4. Cascading¶

2.2.5. Anàlisi¶

3. Classificació seqüencial i paral·lela amb xarxes neuronals NNs (2.5 punts)¶

Objectiu:¶

Tools i Llibreries:¶

En cas de NO utilitzar Colab: Configurar un entorn virtual (opcional però recomanat)¶

Instal·lar TensorFlow¶

Verificació de la instal·lació¶

3.1. Entorn de configuració¶

3.2 Codificación One-Hot de la variable objetivo:¶

3.3. Model de xarxa neuronal seqüencial (1 pt)¶

Arquitectura model:¶

Compilació de models:¶

Entrenant el model:¶

3.4. Model de xarxa neuronal paral·lela (1 pt)¶

Descripció detallada del codi:¶

3.5. Avaluació i anàlisi (0.5 punts)¶

Anàlisi¶

Sequential¶

4. Conclusions (1 punt)¶